GPTZero Performance in Identifying Artificial Intelligence-Generated Medical Texts: A Preliminary Study
10.3346/jkms.2023.38.e319
- Author:
Farrokh HABIBZADEH
1
Author Information
1. Past President, World Association of Medical Editors (WAME) Editorial Consultant, The Lancet Associate Editor, Frontiers in Epidemiology
- Publication Type:Original Article
- From:Journal of Korean Medical Science
2023;38(38):e319-
- CountryRepublic of Korea
- Language:English
-
Abstract:
Background:With emergence of chatbots to help authors with scientific writings, editors should have tools to identify artificial intelligence-generated texts. GPTZero is among the first websites that has sought media attention claiming to differentiate machine-generated from human-written texts.
Methods:Using 20 text pieces generated by ChatGPT in response to arbitrary questions on various topics in medicine and 30 pieces chosen from previously published medical articles, the performance of GPTZero was assessed.
Results:GPTZero had a sensitivity of 0.65 (95% confidence interval, 0.41–0.85); specificity, 0.90 (0.73–0.98); accuracy, 0.80 (0.66–0.90); and positive and negative likelihood ratios, 6.5 (2.1–19.9) and 0.4 (0.2–0.7), respectively.
Conclusion:GPTZero has a low false-positive (classifying a human-written text as machinegenerated) and a high false-negative rate (classifying a machine-generated text as human-written).
